Infusing theory into deep learning for interpretable reactivity prediction

Wang, Shih-Han; Pillai, Hemanth Somarajan; Wang, Siwen; Achenie, Luke E.K.; Xin, Hongliang

doi:10.1038/s41467-021-25639-8

Cited by 62 publications

(70 citation statements)

References 74 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…They use DOE to evaluate activation functions and neurons on each layer to optimize the neural network. In their recent study, Wang et al 110 design their theory‐infused neural networks based on adsorption energy principles for interpretable reactivity prediction. The use of the novel neural differential equation 111 to solve a first‐principles dynamic system represents a hybrid SGML approach, where the architecture of ML model is influenced by the system and finds applications in continuous time series models and scalable normalizing flows.…”

Section: Science Compliments MLmentioning

confidence: 99%

A hybrid science‐guided machine learning approach for modeling chemical processes: A review

Sharma

Liu

2022

AIChE Journal

View full text Add to dashboard Cite

This study presents a broad perspective of hybrid process modeling combining the scientific knowledge and data analytics in bioprocessing and chemical engineering with a science-guided machine learning (SGML) approach. We divide the approach into two major categories: ML complements science, and science complements ML. We review the literature relating to the hybrid SGML approach, and propose a systematic classification of hybrid SGML models. For applying ML to improve science-based models, we present expositions of direct serial and parallel hybrid modeling and their combinations, inverse modeling, reduced-order modeling, quantifying uncertainty in the process and even discovering governing equations of the process model. For applying scientific principles to improve ML models, we discuss the science-guided design, learning and refinement. For each subcategory, we identify its requirements, strengths, and limitations, together with their published and potential applications. We also present several examples to illustrate different hybrid SGML methodologies for modeling chemical processes.

show abstract

Section: Science Compliments MLmentioning

confidence: 99%

A hybrid science‐guided machine learning approach for modeling chemical processes: A review

Sharma

Liu

2022

AIChE Journal

View full text Add to dashboard Cite

show abstract

“…[10,27] A similar approach has been taken in most other works targeting simple adsorbates. [11,14,48,49] In the present work we instead fit a single model to all adsorbates, and the different adsorbates are then instead distinguished from each other via adsorbate-specific features such as HOMO/LUMO energy levels.…”

Section: Further Details On ML Modelsmentioning

confidence: 99%

“…Machine learning (ML) models have already shown their potential for replacing expensive DFT calculations in order to tackle the screening of large materials spaces for accelerated catalyst discovery. [8][9][10][11][12] However, most works so far have been limited in scope to the consideration of atoms or small molecules with mono-dentate adsorption motifs. For these simple species, models now routinely achieve the prediction of adsorption enthalpies with a root-mean-square-error (RMSE) around 0.1-0.2 eV, which is then comparable to the intrinsic DFT accuracy.…”

mentioning

confidence: 99%

Predicting binding motifs of complex adsorbates using machine learning with a physics-inspired graph representation

Xu,

Reuter,

Andersen

2022

Preprint

View full text Add to dashboard Cite

Computational screening in heterogeneous catalysis relies increasingly on machine learning models for predicting key input parameters due to the high cost of computing these directly using firstprinciples methods. This becomes especially relevant when considering complex materials spaces, e.g. alloys, or complex reaction mechanisms with adsorbates that may exhibit bi-or higher-dentate adsorption motifs. Here we present a data-efficient approach to the prediction of binding motifs and associated adsorption enthalpies of complex adsorbates at transition metals (TMs) and their alloys based on a customized Wasserstein Weisfeiler-Lehman graph kernel and Gaussian Process Regression. The model shows good predictive performance, not only for the elemental TMs on which it was trained, but also for an alloy based on these TMs. Furthermore, incorporation of minimal new training data allows for predicting an out-of-domain TM. We believe the model may be useful in active learning approaches, for which we present an ensemble uncertainty estimation approach.

show abstract

“…Bayeschem prediction performance is, however, compromised by its interpretability. The theory-infused neural network (TinNet) was recently developed that infuses the d -band theory of chemisorption into deep learning networks to predict reactivity properties of transition-metal surfaces. As shown in Figure , the TinNet framework contains two sequential components: a regression module and a theory module.…”

mentioning

confidence: 99%

“…(c) TinNet shows coupling integral squared ( V ad 2 ) for 3σ, 1π, and 4σ* orbitals linearly correlates with the corresponding orbital hybridization energy. Adapted from ref . Copyright 2021, Wang et al…”

mentioning

confidence: 99%